Three new sites made 215,128 template pages that Perplexity cited 181 times

Trellner Research published a report on 2 September that measures how much of Perplexity's software recommendations come from three coordinated, newly registered sites that have between them produced 215,128 "best of" category pages. The report is filed as TR-2026-009 and reached 140 points on Hacker News in its first three hours.

The three sites

Trellner names wifitalents.com with 105,541 URLs and 72,713 pages of the form /best/-software/, worldmetrics.org with 103,578 URLs and 70,731 such pages, and gitnux.org with 107,083 URLs and 71,684 such pages. The three share Cloudflare nameservers, use matching templates, and title their pages "Facts and Grounding Page", a phrase written for a retrieval system rather than a human reader. All three domains were registered between December 2023 and May 2024.

How the measurement was done

The report ran 760 calls against Perplexity's sonar and sonar-pro models across 380 software categories, asking for a ranked top five with the official homepage domain of each product. Every domain Perplexity returned in a citation was cross-checked against Tranco's top one million list and the Internet Archive.

Of 7,534 citations gathered in the sample, 23.4 percent pointed to domains outside the top one million, and 59.8 percent ranked worse than the hundred thousandth position on Tranco. Within that group, the three sites contributed 181 citations across 41 of the 380 categories. Trellner rounds this to 2.4 percent of Perplexity's citations in the sample.

What the sites are

None of the three has a long human history that would explain why an answer engine trusts them. They are sites that mass-produce template pages, and their whole purpose is to cover every category so that a retrieval query for "best software" returns a machine-readable page. The consistency across the three domains suggests the same operator behind all of them.

Volume and structure win over reputation in citations

The GEO conversation for the last year has been about how to earn citations from ChatGPT, Perplexity, Google's AI Overviews and Claude. This report is a reminder that citation choices reward structure and volume more than reputation, and that a competitor with a template and a category list is already ahead of a product team that publishes only one main page per product.

Two practical lessons for a product team. First, if you sell software and Perplexity is on your dashboard, add category-page coverage to the audit. A single well-written homepage will lose to a large set of thin category pages that mention your competitor by name. The fix is publishing your own comparison and category content, sourced properly, so that a retrieval query for the category returns a page the answer engine can quote.

Second, if you run search or brand-visibility measurement, treat citation counts on their own with care. A rise in Perplexity mentions may be a rise in real coverage, or it may be one operator publishing another 70,000 templated pages. The check is to look at the domain composition of the citations, and Trellner's method of cross-referencing Tranco and the Internet Archive is a reasonable starting point.

The Trellner report does not name the operator, does not measure ChatGPT, Google AI Overviews or Claude, and takes 380 categories out of a much larger space. Treat the 2.4 percent figure as a sign of the problem rather than its full size.

Source

• Manufactured Sources Behind AI Recommendations, TR-2026-009, Trellner Research, 2 September 2026
• Hacker News discussion